Hydraulic systems are used where relatively high forces and torques are required, or when high response speeds are necessary. They are used in industrial systems, robotics, movement simulators, automated plants, ore exploration, presses, and especially in fatigue testing systems. Fatigue tests are usually performed on servo-hydraulic systems, in order to predict the behavior of materials and their life in service. Fatigue tests are almost always independent of the loading frequency. For a given material and magnitudes of alternate and mean stresses, the fatigue life depends essentially on the number of applied load cycles on the tested material. For this reason, working with the material testing machine at high frequencies brings the advantages of reduction in time and cost, without altering the results. The application of the load can be repeated millions of times, in frequencies of up to one hundred times per second for metals, or even more. To achieve such frequencies, relatively high for a fatigue test, it is necessary to use an efficient control system. In this thesis, learning control techniques are developed and applied to a materials testing machine, allowing the application of constant or variable amplitude loads in high frequency. The proposed methodology consists of implementing a bang-bang type control, restricting the system servo-valve to always work at its extreme limits of operation, i.e., always keeping it completely open in one or the other direction. Due to the system dynamics, the reversion instant must happen before achieving the peaks and valleys of desired force (or stress, strain, etc.). The reversion instant is a parameter that depends on several factors, such as the alternate and mean loading components. It is also influenced by dead zones caused, e.g., by the slack in the mounting between a CTS specimen and the machine pins. As the servo-valve works in its limits of operation, the learning algorithm tries to obtain the optimal instants for the reversions, associating them to a non dimensional variable with values between 0 and 1, stored in specific tables. The learning law constantly updates the values of the table during the execution of the tests, improving the system response. In this work, the dynamic modeling of a servo-hydraulic machine is presented, together with its control scheme. Simulations are performed to compare results from PID and learning controls. The experimental validation is made using a servohydraulic testing machine. For this purpose, real time control software is developed and implemented in a CompactRIO computational system. The results demonstrate the efficiency of the proposed methodology.